The Global Flood Monitoring (GFM) Product of the CEMS is an automated, global, flood monitoring system, which complements the existing CEMS components for flood early warning and on-demand mapping, by providing a continuous (i.e. all-weather, day-and-night), systematic monitoring of all major global flood events, in near real-time (NRT), based on the latest Sentinel-1 SAR satellite images.
For each newly acquired Sentinel-1 SAR satellite image, the GFM Product provides 10 output layers of worldwide flood-related information, which are shown in the dedicated section. Central to the GFM Product are three state-of-the-art algorithms for the SAR-based detection and delineation of flooded areas, which were developed by members of the GFM consortium.
In order to ensure optimal accuracy of the derived flood (and water) extent maps, and to build a high degree of redundancy into the production service, the GFM Product deploys the three state-of-the-art flood mapping algorithms in an “ensemble” approach, whereby an area (i.e. grid-cell) is considered to be flooded if:
Note that in a pre-operational version of the GFM Product, a slightly different ensemble approach was used: in those exceptional cases when only two of the three algorithms produce a result, the classification result of the algorithm with the lower classification uncertainty was selected.
In order to optimize further the quality of the results of the GFM Product, an Exclusion Mask is used to exclude those areas where SAR-based water (and flood) detection is not technically feasible. The GFM Exclusion Mask is created by combining global information layers delineating the following ground surface characteristics:
As specified in the Technical Specifications for the GFM Product (European Commission, 2020), as well as the 10 output layers of flood-related information, the GFM Product can also generate a processed archive of worldwide observed floods and water bodies, from 1 January 2015 until recently.
| # | GFM PRODUCT OUTPUT LAYER | DESCRIPTION AND NOTES |
| 1 | Observed flood Extent | Flooded areas mapped by applying the GFM ensemble flood mapping algorithm to the latest Sentinel-1 images of SAR backscatter intensity. |
| 2 | Observed water extent | Open and calm water mapped as the union of the observed flood extent and the reference water mask. |
| 3 | Reference water mask | Normal (i.e. permanent and seasonal) water mapped by applying the GFM ensemble water mapping algorithm to a historical five-year time-series (“data cube”) of Sentinel-1 images of SAR backscatter intensity. |
| 4 | Exclusion mask | Areas where SAR-based water mapping is not technically feasible, due to no sensitivity (e.g. urban areas, dense vegetation), low backscatter (e.g. flat impervious areas, sandy surfaces), topographic distortions, radar shadows, or low coverage of Sentinel-1. |
| 5 | Likelihood values | Estimated likelihood of flood classification, for all areas outside the exclusion mask. |
| 6 | Advisory flags | Flags indicating potential reduced quality of flood mapping, due to prevailing environmental conditions (e.g. wind, ice, snow, dry soil), or degraded input data quality due to signal interference from other SAR missions; |
| 7 | Sentinel-1 footprint and metadata | Image boundaries of the Sentinel-1 data used, and in addition information on the “metadata”, i.e. information on the acquisition parameters of the Sentinel-1 data used. |
| 8 | Sentinel-1 schedule | Next scheduled Sentinel-1 data acquisition. |
| 9 | Affected population | Number of people in flooded areas, mapped by a spatial overlay of observed flood extent and gridded population, from the Copernicus GHSL project. |
| 10 | Affected Landcover | Land cover / use (e.g. artificial surfaces, agricultural areas) in flooded areas, mapped by a spatial overlay of observed flood extent and the Copernicus GLS land cover. |
The GFM Product output layers of global flood-related information, generated in near real-time based on Sentinel-1 SAR satellite imagery.
| FLOOD MAPPING ALGORITHM | MAIN TECHNICAL FEATURES | SCIENTIFIC REFERENCE |
| Algorithm 1 (LIST) |
|
Chini et al. (2017) |
| Algorithm 2 (DLR) |
|
Martinis et al. (2015) |
| Algorithm 3 (TUW) |
|
Bauer-Marschallinger et al. (2022) |
Overview of the three state-of-the-art GFM algorithms for SAR-based flood mapping.
The performance of the GFM service and product delivery is reported on a quarterly basis, using a minimum set of Key Performance Indicators (KPIs) that monitor the service availability, product timeliness, product quality, unique visitors, total visitors, and total downloads (European Commission, 2020). One of the key user requirements for the GFM Product is the timeliness for generating the GFM output layers of worldwide flood information. The experience of the entire GFM consortium is used to ensure a product timeliness less than or equal to 8 hours, for all Sentinel-1 GRD scenes that become available on the ESA data hubs within 5 hours from acquisition. Some of the practical measures that are used to ensure the product timeliness of the GFM Product, are listed in the table below.
|
# |
DESCRIPTION OF MEASURE |
|
1 |
A redundant Sentinel data access via multi-access point data downloads and parallelization of downloads for an optimised bandwidth utilisation via the hubwatcher software. |
|
2 |
Automatic retrieval of Sentinel-1 GRD data with the hubwatcher in 96.3 minutes (on average) following ESA publication for general availability. |
|
3 |
Pre-processing of Sentinel-1 GRD data products and inclusion into the data cube format within 10-20 minutes per scene. |
|
4 |
Fast data access to the data cube with IO speeds of 15 gibibytes (gib) or 120 gigabytes (GB) per second. |
|
5 |
Generation of the “ensemble” GFM Product using the three parallel flood mapping algorithms consumes up to 30 minutes, with the final flood extent maps and uncertainty values being calculated within a matter of seconds. |
|
6 |
Total lead-time of the GFM Product is 60 minutes, to complete the entire operational NRT workflow from image acquisition to delivery. |
The guaranteed service availability of the GFM Product is facilitated through EODC’s expertise in Sentinel-1 processing operations (begun in 2014), which are regularly assessed by the ESA mission reports. These lessons learned have been used to implement a completely independent IT infrastructure for the CEMS GFM system operations. The operations are also organized through redundant 24/7 production coordinator staff availabilities, such that a 99% service availability is guaranteed.
According to the Technical Specifications (European Commission, 2020), the quality (i.e. thematic accuracy) of the GFM Product’s main output layers Observed Flood Extent and Observed Water Extent, should reach a Critical Success Index (CSI) target threshold of 70-80%. Due to the highly advanced “ensemble” approach integrating the three leading edge GFM flood mapping algorithms, and depending on the quality of input data, a flood and water delineation accuracy which reaches the target threshold is expected. Based on multiple test-cases, the GFM Product usually achieves a satisfactory overall accuracy for both output layers.